agentleFS
carlosbuelvas1carlosbuelvas1/openclaw-skills/thumbnail-creator/SKILL.md

thumbnail-creator

Skill0 starsChanged 6 months agoMITView on GitHub

Use when Charlie wants YouTube or X thumbnails created, improved, or researched from high-performing references using image_generate.

4 other projects use a version of this.

Adopted by

Most recently taken up by

The file

Show its rules and text

We split this skill into the rules it gives an agent

19 rules. 0 are common, 16 are distinctive, and 3 are very specific to this project.

  • 0

    Common

    Most of the 4 edited versions we read keep these.

  • 16

    Distinctive

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use this skill when Charlie asks to:

    line 9
  2. Distinctive:

    Create a YouTube thumbnail, X post image, cover, or content visual.

    line 10
  3. Very specific:

    Before generating ANY thumbnail, read ~/.openclaw/workspace/knowledge/thumbnail-style.md.

    line 16
  4. Distinctive:

    Every concept must align with this style guide.

    line 18
  5. Distinctive:

    Do not generate from taste alone.

    line 21
  6. Distinctive:

    If the user gives no topic, video title, niche, creator, or source, ask one concise question.

    line 23
  7. Distinctive:

    Use available tools before creating.

    line 40
  8. Very specific:

    Read references/scoring-and-prompts.md when doing a full research-backed thumbnail job.

    line 63
  9. Distinctive:

    If subscriber count, view count, or engagement count is missing, label the score as qualitative.

    line 73
  10. Distinctive:

    When the image tool is available, call:

    line 98
  11. Distinctive:

    If Charlie provides a face, logo, screenshot, or prior thumbnail, use it as a reference image when available.

    line 107
  12. Very specific:

    If image_generate is unavailable, output the exact tool-ready prompt package and say the image tool is not currently exposed.

    line 109
  13. Distinctive:

    Use this format for a full thumbnail job:

    line 113
  14. Distinctive:

    Do not say "viral" unless metrics prove it.

    line 151
  15. Distinctive:

    Do not fabricate views, subscribers, likes, thumbnails, or links.

    line 152
  16. Distinctive:

    Do not use tiny text, cluttered backgrounds, or more than one main idea.

    line 153
  17. Distinctive:

    Do not create generic AI-gloss thumbnails with floating icons and no tension.

    line 154
  18. Distinctive:

    Do not copy a competitor thumbnail exactly.

    line 155
  19. Distinctive:

    Do not publish or post externally without Charlie's explicit approval.

    line 156

“Common” is measured: kept by at least 6 in 10 of the 4 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Thumbnail Creator

When to Use

Use this skill when Charlie asks to:

  • Create a YouTube thumbnail, X post image, cover, or content visual.
  • Improve a thumbnail concept, hook, title/thumbnail pairing, or visual packaging.
  • Research what thumbnails are working for a topic, creator, niche, or competitor.
  • Generate image prompts/assets with the ChatGPT/OpenAI image tool.

Style Memory (MANDATORY)

Before generating ANY thumbnail, read ~/.openclaw/workspace/knowledge/thumbnail-style.md. This file contains Charlie's proven preferences, reference creators, DO/DON'T rules, and approval history. Every concept must align with this style guide. After Charlie approves or rejects a thumbnail, update the "Approved Styles" or "Rejected Styles" section in that file — this is how the system learns.

Core Rule

Do not generate from taste alone. Build from evidence AND the style memory file, then create.

If the user gives no topic, video title, niche, creator, or source, ask one concise question. Otherwise make a reasonable assumption and label it.

Workflow

Step 1 - Define The Packaging Target

Capture:

  • Platform: YouTube thumbnail, YouTube Short cover, X image, or multi-platform.
  • Topic/title/hook.
  • Audience and emotion to trigger.
  • Desired face/subject/brand assets if any.
  • Output count and aspect ratio.

Defaults:

  • YouTube: 16:9, 3840x2160, high contrast, readable at phone size.
  • X: 16:9 or 4:5 depending on post style; avoid tiny text.

Step 2 - Research Winners And Metrics

Use available tools before creating. Metrics are required when the platform/tool exposes them.

YouTube metrics to collect:

  • URL, title, channel, publish date, duration, thumbnail URL.
  • Views.
  • Channel subscribers when available.
  • Likes/comments when available.
  • Calculated metrics: views per subscriber, views per day, comments per 1K views, likes per 1K views.

X metrics to collect:

  • URL, author, post date, media URL/screenshot when available.
  • Views/impressions when available.
  • Likes, reposts, replies, bookmarks when available.
  • Calculated metrics: engagement count, engagement rate, repost rate, reply rate.

Tools:

  • YouTube: use youtube_video_fetch for supplied URLs; use web/search tools when available to discover references.
  • X: use x_user_timeline_fetch or available X tools for creator/topic references.
  • Web/search tools: use only if available. If search is unavailable, ask Charlie for 3-5 reference URLs or creator names.

Collect 5-12 references when possible. Never invent metrics or URLs. If a metric is hidden or unavailable, write unavailable and rank that item with a qualitative caveat.

Step 3 - Score The Pattern

Read references/scoring-and-prompts.md when doing a full research-backed thumbnail job.

Score references by:

  • Outlier strength: views relative to channel/subscriber baseline when available.
  • Engagement strength: likes/comments/reposts per view when available.
  • Recency: newer references matter more for fast-moving niches.
  • Visual clarity: readable at small size.
  • Curiosity gap: makes the viewer need the answer.
  • Topic match: relevant to Charlie's exact idea, not just broadly popular.

If subscriber count, view count, or engagement count is missing, label the score as qualitative. Do not declare a reference the winner unless either metrics support it or you explicitly say it is a qualitative visual pick.

Step 4 - Extract The Winning Formula

Summarize the pattern in operator language:

  • Dominant visual object/person.
  • Emotion on face or subject posture.
  • Text count and exact text style.
  • Color contrast and background.
  • Curiosity mechanism.
  • What to avoid.

Step 5 - Create Concepts

Produce 3 concepts before image generation:

  1. Safe proven winner.
  2. Contrarian/underdog angle.
  3. Weird high-risk hook.

Each concept must include:

  • Thumbnail title.
  • On-image text, max 3-5 words unless Charlie asks otherwise.
  • Composition.
  • Why it should earn the click.
  • Risk.

Step 6 - Generate With image_generate

When the image tool is available, call:

  • Tool: image_generate
  • Model: openai/gpt-image-2
  • YouTube size: 3840x2160
  • Aspect ratio: 16:9
  • Count: 1-4
  • Quality: high
  • Filename: descriptive slug.

If Charlie provides a face, logo, screenshot, or prior thumbnail, use it as a reference image when available. Preserve likeness/brand assets without inventing identity details.

If image_generate is unavailable, output the exact tool-ready prompt package and say the image tool is not currently exposed.

Output Format

Use this format for a full thumbnail job:

## Research Read
- Reference set: [N] sources
- Best metric-backed winner: [title/creator + why]
- Best visual pattern: [one sharp sentence]
- Metrics caveat: [only if needed]

## Metrics Scorecard
| Rank | Source | Views/Impressions | Subs/Followers | Age | Engagement | Calculated Signal | URL |
|---|---|---:|---:|---:|---:|---|---|
| 1 | [title/creator] | [views] | [subs/followers or unavailable] | [days] | [likes/comments/reposts/etc.] | [views/sub, views/day, engagement rate] | [URL] |

## Winning References
1. [Title/creator] - [metric-backed reason] - [URL]
2. ...

## Concepts
### 1. [Concept Name]
- On-image text: "..."
- Composition: ...
- Click mechanism: ...
- Risk: ...

### 2. ...
### 3. ...

## Image Prompt To Run
[Final prompt for image_generate]

## Generated Assets
[Only include after image_generate returns media paths/URLs]

For quick requests, skip the long research report and give 3 concepts plus one final image prompt.

What To Avoid

  • Do not say "viral" unless metrics prove it.
  • Do not fabricate views, subscribers, likes, thumbnails, or links.
  • Do not use tiny text, cluttered backgrounds, or more than one main idea.
  • Do not create generic AI-gloss thumbnails with floating icons and no tension.
  • Do not copy a competitor thumbnail exactly; extract the pattern and make it Charlie's.
  • Do not publish or post externally without Charlie's explicit approval.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.